In order to fully consume renewable energies and schedule demand-side resources more hierarchically, a doubledeck optimal schedule model of micro-grid, which connects to power grids and concludes battery energy storage system (BESS), is proposed. Unlike the original peak-valley time-of-use (TOU) price, in the upper layer optimal schedule, the improved TOU price which takes account into the user's satisfaction can express the adjusted loads accurately and the resulting net loads can be treated as a link between upper and lower layer scheduling. For the reason of having no precise model of BESS, the lower model for the goal of minimising the operation cost is solved by action dependent heuristic dynamic programming algorithm that is not relying on the accurate controlled object model. This algorithm is used to obtain the most optimal performance index function and control strategy by the optimal iterative process, which is based on the back propagation neural network used for evaluating the optimal performance index. Analysis of examples and results has been presented to show the effectiveness of the proposed strategies.